Predictive AI Model Identifies nAMD Treatment Response Biomarkers


A new assessment from ARVO 2022 showed methodology to predict one-year treatment response to treat-and-extend ranibizumab at 1 year.

Predictive AI Model Identifies nAMD Treatment Response Biomarkers

Hrvoje Bogunovic, PhD

An artificial intelligence (AI) method designed to predict visual outcomes via a treat-and-extend (T&E) regimen of anti-VEGF therapy ranibizumab in patients with neovascular age-related macular degeneration (nAMD) showed particular benefit of defined—and easily accessible—biomarkers.

The new data, presented at the Association for Research in Vision and Ophthalmology (ARVO) 2022 Meeting this week, may contribute the developing field of AI and machine learning for optimizing chronic retinal disease management.

Led by Hrvoje Bogunovic, PhD, director of the Christian Doppler Lab for Artificial Intelligence in Retina and associates director of the Laboratory for Ophthalmic Image Analysis at the Medical University of Vienna, a team of investigators developed a predictive model designed to predict one-year visual outcomes in treated patients with nAMD based on measurable, predictive optical coherence tomography (OCT)-based biomarkers.

Their assessment presented at ARVO 2022 observed the predictive model across 550 patient eyes receiving ranibizumab T&E regimen for 1 year.

As Bogunovic and colleagues noted, T&E has become a “dominant treatment regimen” for patients with nAMD prescribed to anti-VEGF therapies. “However, retinal morphology and dynamics predictive of patient outcomes are still unclear,” they wrote.

The team used latent class mixed models to design patient visual acuity trajectories. They analyzed macular OCT volume scans of treatment-naive eyes at baseline and 4 weeks after initial ranibizumab injection via AI. Deep learning was used to quantify patient intraretinal fluid (IRF), sub retinal fluid (SRF), pigment epithelial detachment (PED), and sub retinal hyper reflective material (SHRM).

Early Treatment Diabetic Retinopathy Study (ETDRS) grid-based measurements were used to represent patient retinal morphology and characterize spatial distribution and change post-initial injection.

The patient population was pooled from clinical trial data including 550 patients with as many affected nAMD eyes. Based on the latent class mixed models analysis of visual acuity trajectories, approximately one-fourth (23%) of assessed eyes were considered clinically unresponsive. The AI model helped predict unresponsiveness with an area under the receiver-operating curve (AUROC) of 0.74 (95% CI, 0.71 - 78).

Investigators noted that remaining IRF amount after the first ranibizumab injection at 1 month was the most important predictive biomarker in the treated patient eyes.

“Predictive retinal OCT biomarkers help explain the different functional results after treatment between patients having similar baseline visual acuity initially,” investigators concluded. “The result of this study is a promising step toward AI-based prediction of patient outcomes and may help to further personalize and optimize anti-VEGF therapy in clinical practice.”

The study, "Artificial Intelligence to Identify Retinal OCT Parameters Predictive of Patient Visual Outcomes under a Treat-and-Extend Regimen," was presented at ARVO 2022.

Related Videos
Hong-Uyen Hua, MD; Sruthi Arepalli, MD; and Peter Kaiser, MD
Connective Tissue Disease Brings Dermatology & Rheumatology Together
Kelly Nichols, OD; Laura Periman, MD; and Mile Brujic, OD
© 2024 MJH Life Sciences

All rights reserved.